HFD Final Project – Group 1

Data Specifications:

• Assets: S&P 500 (SP) & NASDAQ (NQ)
  
• Frequency: 1-minute bars
• Period: 2023 Q1 – 2025 Q2
• Trading: Separate signals, portfolio aggregation

Trading Constraints (both):

• Excluded windows: 9:31–9:40, 15:51–16:00
• No trading: 9:31–9:55 (volatility buffer)
• Forced flat: 15:40 onwards
• No overnight positions

Strategy 1: Opening Range Breakout

Key Features:

• Opening range: 09:41–09:55
• Breakout entry with ATR buffer
• Mean-reversion exits
• Low turnover design

Strategy 2: Z Guided Trend

Key Features:

• Trend-following entries (SMA vs SMA_slow level + slope gate)
• z-score entry with z-momentum confirmation (z accelerating into entry)
• Hard constraints: no-trade window, no new entries after 14:30, forced flat from 15:40
• Risk controls: z stop-loss + extreme z stop, cooldown after exits (reduces churn)

3-Way Data Split:

• Training (4 qtrs): 2023_Q1, Q3, 2024_Q2, Q4
• Validation (2 qtrs): 2023_Q4, 2025_Q1
  
• Test (1 qtr): 2025_Q2

Grid Search:

• ATR Window: [15, 30, 60] minutes
• ATR Multiplier: [0.0, 0.05, 0.10, 0.15, 0.20]
• Selection: Based on VALIDATION performance

Optimal Parameters:

# Best parameters (selected on validation)
print("SP: ATR_window=15, ATR_mult=0.0")
print("NQ: ATR_window=30, ATR_mult=0.2")

SP: ATR_window=15, ATR_mult=0.0
NQ: ATR_window=30, ATR_mult=0.2

STAT Metric:

\[\text{STAT} = (\text{Sharpe}_{\text{net}} - 0.5) \times \ln(|\text{Net PnL}/1000|)\]

• Penalizes low Sharpe ratios
• Rewards larger absolute PnL
• Balances risk-adjusted and absolute returns

Methodology:

1. Evaluate all combinations on TRAIN
2. Select best based on VALIDATION
   
3. Final evaluation on TEST (unseen)

Train/Val/Test Split:

• Training (5 qtrs): 2023_Q1, Q3, Q4, 2024_Q4, 2025_Q1
• Validation (1 qtr): 2024_Q2
  
• Test (1 qtr): 2025_Q2 (completely unseen)

Grid Search:

• SMA Window: [240, 360, 480] minutes
  
• Slow Multiplier: [3, 4]
  
• Slope Lookback (m): [5, 10]
  
• Z Entry Threshold: [1.4, 1.6, 1.8, 2.0]
  
• Z Stop Threshold: [3.5, 4.0, 4.5]
  
• Z Stop-Loss (SL) Threshold: [1.0, 1.5, 2.0]
  
• Z Momentum Threshold: [0.05, 0.10, 0.20, 0.30]
  
• Cooldown: [5, 10]
  
• Selection: Based on VALIDATION performance

Optimization Objective:

• We select parameters that maximize Net P&L (transaction costs included).
• This penalizes over-trading automatically via per-trade costs.
• Balances risk-adjusted and absolute returns

Methodology:

1. Evaluate all combinations on TRAIN.
2. Use same quarter combination on next year. (Seasonality)
3. Final evaluation on TEST

Optimal Parameters:

NQ: SMA_win=360, slow_mult=3, slope_m=10, z_entry=1.6, z_stop=3.5, z_sl=2.0, z_mom=0.1, cooldown=5
SP: SMA_win=360, slow_mult=3, slope_m=5, z_entry=1.4, z_stop=3.5, z_sl=2.0, z_mom=0.05, cooldown=5

Strengths:

• Consistent positive returns across most quarters
• Low turnover (controlled transaction costs)
• Clear entry/exit rules
• Adapts to volatility via ATR

Key Observations:

• Strategy performs well in trending markets
• ATR filter effectively reduces false breakouts
• Mean-reversion exit manages risk

Risk Management:

• No overnight positions (flat at 15:40)
• One entry per day (prevents overtrading)
• Volatility-adjusted thresholds
• Excludes unreliable time windows

Overall Portfolio Performance:

overall_stat = orb_port["stat"].sum()
overall_pnl = orb_port["net_cumPnL"].sum()
overall_sharpe = orb_port["netSR"].mean()
print(f"Total Net PnL: ${overall_pnl:,.2f}")
print(f"Total STAT Score: {overall_stat:.4f}")
print(f"Avg Net Sharpe: {overall_sharpe:.3f}")

Total Net PnL: $70,954.54
Total STAT Score: 16.6499
Avg Net Sharpe: 1.287

Strengths:

• Trend-aligned entries via fast/slow SMA filter
• Uses z-momentum confirmation to reduce early/late fades
• Intraday-only design (no overnight exposure)
• Works on both SP and NQ with the same logic

Key Observations:

• Best performance tends to occur in sustained intraday trends
• Z-momentum filter helps avoid entries on flattening/extreme z moves
• Cooldown reduces rapid flip-flopping after exits

Risk Management:

• No trading during 09:31–09:55 (volatility buffer)
• No new entries after 14:30
• Forced flat from 15:40 onwards
• Stop-loss in z-space + extreme z exits
• Transaction costs included in net P&L

Overall Portfolio Performance:

import pandas as pd
from pathlib import Path

BASE_DIR = Path.cwd()
OUT_DIR  = BASE_DIR.parent / "Testing" / "outputs_strategy02"
perf = pd.read_csv(OUT_DIR / "strategy02_perf.csv")

port = perf[perf["asset"]=="PORTFOLIO"]
overall_pnl = port['net_cumPnL'].sum()
overall_sharpe = port['netSR'].mean()
# Calculate STAT for Strategy 2
overall_stat = sum([(sr - 0.5) * np.log(abs(pnl/1000)) if pnl != 0 and np.isfinite(sr) else 0 
                    for sr, pnl in zip(port['netSR'], port['net_cumPnL'])])
print(f"Total Net PnL: ${overall_pnl:,.2f}")
print(f"Total STAT Score: {overall_stat:.4f}")
print(f"Avg Net Sharpe: {overall_sharpe:.3f}")

Total Net PnL: $193,456.15
Total STAT Score: 56.8480
Avg Net Sharpe: 2.833